If you’ve been following the AI and jobs debate, you’ve probably seen the chart Anthropic dropped this month. It’s a two-tone bar graph comparing current “observed exposure” (red) to “theoretical capability” (blue) across 22 job categories. The red bars are modest. The blue ones? They stretch like they’re trying to swallow the whole labor market.
At first glance, it looks like Anthropic is saying LLMs could theoretically handle at least 80 percent of tasks in fields ranging from Arts & Media to Legal, Business & Finance, and even Management. That’s the kind of number that makes you wonder if we should all start retraining for something robots can’t do yet.
But here’s the thing: that blue area isn’t a prediction. It’s not even a current measurement. It’s a collection of educated guesses—some outdated, some wildly speculative—about where AI might improve human productivity. Not replace it. Improve it. That’s a big difference.
Anthropic’s methodology for “theoretical capability” leans heavily on expert surveys and extrapolations from existing benchmarks. The problem is that benchmarks don’t always translate to real-world job performance. An LLM might ace a legal reasoning test but still hallucinate case law in a deposition. It might draft a decent marketing copy but have no clue about brand strategy or client relationships.
I’ve seen this pattern before. Every few years, a new study claims AI will automate half of all jobs. It happened with expert systems in the 80s, with the internet in the 90s, with machine learning in the 2010s. Each time, the reality was messier and slower than the headlines suggested.
What’s more useful here is the red part of the chart—the “observed exposure” data. That’s based on actual usage patterns from Claude users. And those numbers are much more modest. They show where people are actually using LLMs today, not where some hypothetical future model might be.
The blue bars are interesting for R&D planning. They’re not a roadmap for mass unemployment. If you’re worried about your job, look at what’s actually happening, not what a speculative chart implies might happen someday.
Anthropic is a smart company doing important work. But this chart is more marketing than science. The real story is in the gap between red and blue—and how we choose to close it.
Comments (0)
Login Log in to comment.
Be the first to comment!